QwenLM / Qwen-Audio

The official repo of Qwen-Audio (通义千问-Audio) chat & pretrained large audio language model proposed by Alibaba Cloud.
Other
1.49k stars 107 forks source link

训练问题 #7

Closed Mddct closed 11 months ago

Mddct commented 11 months ago

请问,第一阶段训练,通义的embedding是打开的吗?

faychu commented 11 months ago

第一阶段训练时,通义千问LLM的word embedding和对应的output layer参数是freeze的,但是新加入的audio special token的embedding和对应的output layer是unfreeze的。

fzhml commented 5 months ago

那为什么看起来第一阶段训练后的正常token的word embedding和qwen-7B的word embedding不一样呢?是用了非官方release的qwen-7B不